Spread Your Claudes — AI skill for Claude Code

---

The problem

Coding agents like Claude Code re-send the **entire conversation** to the model on every single turn. Read a 600-line file into the session and you keep paying for those 600 lines on **every message after it** — until the session ends.

Long *"read a lot, then act"* sessions burn your usage limit on **carrying the haystack**, not on thinking.

The fix

`spread-your-claudes` is a handful of tiny CLI wrappers. Your agent hands the big read to a **research lane** — a one-shot call to a *different* AI CLI (Gemini, Cursor, Codex, a paid Gemini web plan) — which reads on **its** provider's quota and returns a short, cited answer.

Your session grows by **5 lines** instead of **600**.

  YOU
   │
   ▼
┌────────────────────────────────┐        ← this box is your usage limit.
│  PRIMARY AGENT  (Claude Code)  │          everything in it is re-billed
│  owns the conversation         │          every turn.
│  decides · writes · commits    │
└───────────────┬────────────────┘
                │  "read this haystack. answer sharp. cite file:line."
                ▼
┌────────────────────────────────┐        ← this runs on the LANE's quota.
│  RESEARCH LANE (fire & return) │          it chews 600 lines of repo /
│  gemini · cursor · codex ·     │          logs / docs here — you don't
│  gemini-web